diff options
| author | Denes Balatoni <dbalatoni@programozo.hu> | 2004-01-03 20:14:45 +0000 |
|---|---|---|
| committer | Michael Niedermayer <michaelni@gmx.at> | 2004-01-03 20:14:45 +0000 |
| commit | c953e7976ec71c007bdc0aa0316398541eb260aa (patch) | |
| tree | 6bbc34955d46249af82991b1f3d5c7c977c0cd8b /libavcodec/i386 | |
| parent | 76e9d39297c43f56098b1eb404d0795d9515b279 (diff) | |
| download | ffmpeg-c953e7976ec71c007bdc0aa0316398541eb260aa.tar.gz | |
denoise_dct_sse2() patch by (Balatoni Denes <pnis at coder dot hu>)
Originally committed as revision 2659 to svn://svn.ffmpeg.org/ffmpeg/trunk
Diffstat (limited to 'libavcodec/i386')
| -rw-r--r-- | libavcodec/i386/mpegvideo_mmx.c | 60 |
1 files changed, 59 insertions, 1 deletions
diff --git a/libavcodec/i386/mpegvideo_mmx.c b/libavcodec/i386/mpegvideo_mmx.c index b6f249ee79..daffd1d8d6 100644 --- a/libavcodec/i386/mpegvideo_mmx.c +++ b/libavcodec/i386/mpegvideo_mmx.c @@ -617,6 +617,60 @@ static void denoise_dct_mmx(MpegEncContext *s, DCTELEM *block){ ); } +static void denoise_dct_sse2(MpegEncContext *s, DCTELEM *block){ + const int intra= s->mb_intra; + int *sum= s->dct_error_sum[intra]; + uint16_t *offset= s->dct_offset[intra]; + + s->dct_count[intra]++; + + asm volatile( + "pxor %%xmm7, %%xmm7 \n\t" + "1: \n\t" + "pxor %%xmm0, %%xmm0 \n\t" + "pxor %%xmm1, %%xmm1 \n\t" + "movdqa (%0), %%xmm2 \n\t" + "movdqa 16(%0), %%xmm3 \n\t" + "pcmpgtw %%xmm2, %%xmm0 \n\t" + "pcmpgtw %%xmm3, %%xmm1 \n\t" + "pxor %%xmm0, %%xmm2 \n\t" + "pxor %%xmm1, %%xmm3 \n\t" + "psubw %%xmm0, %%xmm2 \n\t" + "psubw %%xmm1, %%xmm3 \n\t" + "movdqa %%xmm2, %%xmm4 \n\t" + "movdqa %%xmm3, %%xmm5 \n\t" + "psubusw (%2), %%xmm2 \n\t" + "psubusw 16(%2), %%xmm3 \n\t" + "pxor %%xmm0, %%xmm2 \n\t" + "pxor %%xmm1, %%xmm3 \n\t" + "psubw %%xmm0, %%xmm2 \n\t" + "psubw %%xmm1, %%xmm3 \n\t" + "movdqa %%xmm2, (%0) \n\t" + "movdqa %%xmm3, 16(%0) \n\t" + "movdqa %%xmm4, %%xmm6 \n\t" + "movdqa %%xmm5, %%xmm0 \n\t" + "punpcklwd %%xmm7, %%xmm4 \n\t" + "punpckhwd %%xmm7, %%xmm6 \n\t" + "punpcklwd %%xmm7, %%xmm5 \n\t" + "punpckhwd %%xmm7, %%xmm0 \n\t" + "paddd (%1), %%xmm4 \n\t" + "paddd 16(%1), %%xmm6 \n\t" + "paddd 32(%1), %%xmm5 \n\t" + "paddd 48(%1), %%xmm0 \n\t" + "movdqa %%xmm4, (%1) \n\t" + "movdqa %%xmm6, 16(%1) \n\t" + "movdqa %%xmm5, 32(%1) \n\t" + "movdqa %%xmm0, 48(%1) \n\t" + "addl $32, %0 \n\t" + "addl $64, %1 \n\t" + "addl $32, %2 \n\t" + "cmpl %3, %0 \n\t" + " jb 1b \n\t" + : "+r" (block), "+r" (sum), "+r" (offset) + : "r"(block+64) + ); +} + #undef HAVE_MMX2 #define RENAME(a) a ## _MMX #define RENAMEl(a) a ## _mmx @@ -643,7 +697,11 @@ void MPV_common_init_mmx(MpegEncContext *s) draw_edges = draw_edges_mmx; - s->denoise_dct= denoise_dct_mmx; + if (mm_flags & MM_SSE2) { + s->denoise_dct= denoise_dct_sse2; + } else { + s->denoise_dct= denoise_dct_mmx; + } if(dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX){ if(mm_flags & MM_MMXEXT){ |